自然最近邻居的特性分析与应用研究

项目名称： 自然最近邻居的特性分析与应用研究

项目编号： No.61272194

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 朱庆生

作者单位： 重庆大学

项目金额： 80万元

中文摘要： 最近邻居的概念及邻居搜索算法是数据挖掘、模式识别、图像处理、智能推荐等领域的一个基础科学问题，本项目围绕这一基础问题，探索新的思维方法，其研究具有十分重要的学术意义和应用价值。项目重点针对基于最近邻居的流形学习和聚类学习算法中"如何选择邻域参数?"这一公开问题和"如何度量空间结构的相似性？"等热点问题，通过引入新的概念"自然最近邻居"，构造新的工具"自适应最近邻域图"，力图改变传统的依赖于参数形式的k最近邻居或ε最近邻域的格局，探索出新的解决方案。 项目重点研究内容包括：自然最近邻居的理论模型、自然最近邻居的高效搜索算法、自然最近邻居和自适应最近邻域图的特征分析、自然最近邻居分类器设计、自然最近邻居分布的信息熵评估等。项目力争取得原创性高水平应用基础研究成果，申请发明专利2-3项，发表高水平论文10篇，培养博士3-4人。

中文关键词： k-最近邻居；自然最近邻居；自然邻居邻域图；流行学习；聚类分析

英文摘要： The concept and the algorithms about the Nearest Neighbor are a foundation scientific issue in many research fields such as Data Mining, Pattern Recognition, Image Processing, and Intelligent Recommendation. This project will explore courageously creative method on this issue, which has important academic significances and application values. The project focuses on a new concept called as "Natural Nearest Neighbor (3N)", which is a scale-free nearest neighbor for trying to solve more effectively the open problem of "selection of neighborhood size" in the manifold learning algorithms and the hot problem of "similarity measurement of dataset space structure" in the clustering learning algorithms. Based on a new graph tool called as "Adaptive Natural Nearest Neighborhood Graph (A3NG)" constructed by relationship among natural neighbors, we strive to propose new manifold learning algorithms without parameter and prove their validity in application fields. The main research contents are the theory modeling for 3N, the neighbor searching algorithm, the characteristics analysis of A3NG, the design and implementation of 3N Classifier, and the information entropy of 3N distribution. The goals of project achieve creativity research results in publishing 10 scientific papers, applying 2-3 national patents for invention of

英文关键词： k-Nearest Neighbor；Natural Nearest Neighbor；Natural Neighbor Neighborhood Graphs；Manifold Learning；Clustering Analysis